xref: /openbmc/linux/net/openvswitch/actions.c (revision 95e9fd10) 
1 /*
2  * Copyright (c) 2007-2012 Nicira, Inc.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16  * 02110-1301, USA
17  */
18 
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 
21 #include <linux/skbuff.h>
22 #include <linux/in.h>
23 #include <linux/ip.h>
24 #include <linux/openvswitch.h>
25 #include <linux/tcp.h>
26 #include <linux/udp.h>
27 #include <linux/in6.h>
28 #include <linux/if_arp.h>
29 #include <linux/if_vlan.h>
30 #include <net/ip.h>
31 #include <net/checksum.h>
32 #include <net/dsfield.h>
33 
34 #include "datapath.h"
35 #include "vport.h"
36 
37 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
38 			const struct nlattr *attr, int len, bool keep_skb);
39 
40 static int make_writable(struct sk_buff *skb, int write_len)
41 {
42 	if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
43 		return 0;
44 
45 	return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
46 }
47 
48 /* remove VLAN header from packet and update csum accrodingly. */
49 static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
50 {
51 	struct vlan_hdr *vhdr;
52 	int err;
53 
54 	err = make_writable(skb, VLAN_ETH_HLEN);
55 	if (unlikely(err))
56 		return err;
57 
58 	if (skb->ip_summed == CHECKSUM_COMPLETE)
59 		skb->csum = csum_sub(skb->csum, csum_partial(skb->data
60 					+ ETH_HLEN, VLAN_HLEN, 0));
61 
62 	vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
63 	*current_tci = vhdr->h_vlan_TCI;
64 
65 	memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
66 	__skb_pull(skb, VLAN_HLEN);
67 
68 	vlan_set_encap_proto(skb, vhdr);
69 	skb->mac_header += VLAN_HLEN;
70 	skb_reset_mac_len(skb);
71 
72 	return 0;
73 }
74 
75 static int pop_vlan(struct sk_buff *skb)
76 {
77 	__be16 tci;
78 	int err;
79 
80 	if (likely(vlan_tx_tag_present(skb))) {
81 		skb->vlan_tci = 0;
82 	} else {
83 		if (unlikely(skb->protocol != htons(ETH_P_8021Q) ||
84 			     skb->len < VLAN_ETH_HLEN))
85 			return 0;
86 
87 		err = __pop_vlan_tci(skb, &tci);
88 		if (err)
89 			return err;
90 	}
91 	/* move next vlan tag to hw accel tag */
92 	if (likely(skb->protocol != htons(ETH_P_8021Q) ||
93 		   skb->len < VLAN_ETH_HLEN))
94 		return 0;
95 
96 	err = __pop_vlan_tci(skb, &tci);
97 	if (unlikely(err))
98 		return err;
99 
100 	__vlan_hwaccel_put_tag(skb, ntohs(tci));
101 	return 0;
102 }
103 
104 static int push_vlan(struct sk_buff *skb, const struct ovs_action_push_vlan *vlan)
105 {
106 	if (unlikely(vlan_tx_tag_present(skb))) {
107 		u16 current_tag;
108 
109 		/* push down current VLAN tag */
110 		current_tag = vlan_tx_tag_get(skb);
111 
112 		if (!__vlan_put_tag(skb, current_tag))
113 			return -ENOMEM;
114 
115 		if (skb->ip_summed == CHECKSUM_COMPLETE)
116 			skb->csum = csum_add(skb->csum, csum_partial(skb->data
117 					+ ETH_HLEN, VLAN_HLEN, 0));
118 
119 	}
120 	__vlan_hwaccel_put_tag(skb, ntohs(vlan->vlan_tci) & ~VLAN_TAG_PRESENT);
121 	return 0;
122 }
123 
124 static int set_eth_addr(struct sk_buff *skb,
125 			const struct ovs_key_ethernet *eth_key)
126 {
127 	int err;
128 	err = make_writable(skb, ETH_HLEN);
129 	if (unlikely(err))
130 		return err;
131 
132 	memcpy(eth_hdr(skb)->h_source, eth_key->eth_src, ETH_ALEN);
133 	memcpy(eth_hdr(skb)->h_dest, eth_key->eth_dst, ETH_ALEN);
134 
135 	return 0;
136 }
137 
138 static void set_ip_addr(struct sk_buff *skb, struct iphdr *nh,
139 				__be32 *addr, __be32 new_addr)
140 {
141 	int transport_len = skb->len - skb_transport_offset(skb);
142 
143 	if (nh->protocol == IPPROTO_TCP) {
144 		if (likely(transport_len >= sizeof(struct tcphdr)))
145 			inet_proto_csum_replace4(&tcp_hdr(skb)->check, skb,
146 						 *addr, new_addr, 1);
147 	} else if (nh->protocol == IPPROTO_UDP) {
148 		if (likely(transport_len >= sizeof(struct udphdr))) {
149 			struct udphdr *uh = udp_hdr(skb);
150 
151 			if (uh->check || skb->ip_summed == CHECKSUM_PARTIAL) {
152 				inet_proto_csum_replace4(&uh->check, skb,
153 							 *addr, new_addr, 1);
154 				if (!uh->check)
155 					uh->check = CSUM_MANGLED_0;
156 			}
157 		}
158 	}
159 
160 	csum_replace4(&nh->check, *addr, new_addr);
161 	skb->rxhash = 0;
162 	*addr = new_addr;
163 }
164 
165 static void set_ip_ttl(struct sk_buff *skb, struct iphdr *nh, u8 new_ttl)
166 {
167 	csum_replace2(&nh->check, htons(nh->ttl << 8), htons(new_ttl << 8));
168 	nh->ttl = new_ttl;
169 }
170 
171 static int set_ipv4(struct sk_buff *skb, const struct ovs_key_ipv4 *ipv4_key)
172 {
173 	struct iphdr *nh;
174 	int err;
175 
176 	err = make_writable(skb, skb_network_offset(skb) +
177 				 sizeof(struct iphdr));
178 	if (unlikely(err))
179 		return err;
180 
181 	nh = ip_hdr(skb);
182 
183 	if (ipv4_key->ipv4_src != nh->saddr)
184 		set_ip_addr(skb, nh, &nh->saddr, ipv4_key->ipv4_src);
185 
186 	if (ipv4_key->ipv4_dst != nh->daddr)
187 		set_ip_addr(skb, nh, &nh->daddr, ipv4_key->ipv4_dst);
188 
189 	if (ipv4_key->ipv4_tos != nh->tos)
190 		ipv4_change_dsfield(nh, 0, ipv4_key->ipv4_tos);
191 
192 	if (ipv4_key->ipv4_ttl != nh->ttl)
193 		set_ip_ttl(skb, nh, ipv4_key->ipv4_ttl);
194 
195 	return 0;
196 }
197 
198 /* Must follow make_writable() since that can move the skb data. */
199 static void set_tp_port(struct sk_buff *skb, __be16 *port,
200 			 __be16 new_port, __sum16 *check)
201 {
202 	inet_proto_csum_replace2(check, skb, *port, new_port, 0);
203 	*port = new_port;
204 	skb->rxhash = 0;
205 }
206 
207 static void set_udp_port(struct sk_buff *skb, __be16 *port, __be16 new_port)
208 {
209 	struct udphdr *uh = udp_hdr(skb);
210 
211 	if (uh->check && skb->ip_summed != CHECKSUM_PARTIAL) {
212 		set_tp_port(skb, port, new_port, &uh->check);
213 
214 		if (!uh->check)
215 			uh->check = CSUM_MANGLED_0;
216 	} else {
217 		*port = new_port;
218 		skb->rxhash = 0;
219 	}
220 }
221 
222 static int set_udp(struct sk_buff *skb, const struct ovs_key_udp *udp_port_key)
223 {
224 	struct udphdr *uh;
225 	int err;
226 
227 	err = make_writable(skb, skb_transport_offset(skb) +
228 				 sizeof(struct udphdr));
229 	if (unlikely(err))
230 		return err;
231 
232 	uh = udp_hdr(skb);
233 	if (udp_port_key->udp_src != uh->source)
234 		set_udp_port(skb, &uh->source, udp_port_key->udp_src);
235 
236 	if (udp_port_key->udp_dst != uh->dest)
237 		set_udp_port(skb, &uh->dest, udp_port_key->udp_dst);
238 
239 	return 0;
240 }
241 
242 static int set_tcp(struct sk_buff *skb, const struct ovs_key_tcp *tcp_port_key)
243 {
244 	struct tcphdr *th;
245 	int err;
246 
247 	err = make_writable(skb, skb_transport_offset(skb) +
248 				 sizeof(struct tcphdr));
249 	if (unlikely(err))
250 		return err;
251 
252 	th = tcp_hdr(skb);
253 	if (tcp_port_key->tcp_src != th->source)
254 		set_tp_port(skb, &th->source, tcp_port_key->tcp_src, &th->check);
255 
256 	if (tcp_port_key->tcp_dst != th->dest)
257 		set_tp_port(skb, &th->dest, tcp_port_key->tcp_dst, &th->check);
258 
259 	return 0;
260 }
261 
262 static int do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
263 {
264 	struct vport *vport;
265 
266 	if (unlikely(!skb))
267 		return -ENOMEM;
268 
269 	vport = rcu_dereference(dp->ports[out_port]);
270 	if (unlikely(!vport)) {
271 		kfree_skb(skb);
272 		return -ENODEV;
273 	}
274 
275 	ovs_vport_send(vport, skb);
276 	return 0;
277 }
278 
279 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
280 			    const struct nlattr *attr)
281 {
282 	struct dp_upcall_info upcall;
283 	const struct nlattr *a;
284 	int rem;
285 
286 	upcall.cmd = OVS_PACKET_CMD_ACTION;
287 	upcall.key = &OVS_CB(skb)->flow->key;
288 	upcall.userdata = NULL;
289 	upcall.pid = 0;
290 
291 	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
292 		 a = nla_next(a, &rem)) {
293 		switch (nla_type(a)) {
294 		case OVS_USERSPACE_ATTR_USERDATA:
295 			upcall.userdata = a;
296 			break;
297 
298 		case OVS_USERSPACE_ATTR_PID:
299 			upcall.pid = nla_get_u32(a);
300 			break;
301 		}
302 	}
303 
304 	return ovs_dp_upcall(dp, skb, &upcall);
305 }
306 
307 static int sample(struct datapath *dp, struct sk_buff *skb,
308 		  const struct nlattr *attr)
309 {
310 	const struct nlattr *acts_list = NULL;
311 	const struct nlattr *a;
312 	int rem;
313 
314 	for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
315 		 a = nla_next(a, &rem)) {
316 		switch (nla_type(a)) {
317 		case OVS_SAMPLE_ATTR_PROBABILITY:
318 			if (net_random() >= nla_get_u32(a))
319 				return 0;
320 			break;
321 
322 		case OVS_SAMPLE_ATTR_ACTIONS:
323 			acts_list = a;
324 			break;
325 		}
326 	}
327 
328 	return do_execute_actions(dp, skb, nla_data(acts_list),
329 						 nla_len(acts_list), true);
330 }
331 
332 static int execute_set_action(struct sk_buff *skb,
333 				 const struct nlattr *nested_attr)
334 {
335 	int err = 0;
336 
337 	switch (nla_type(nested_attr)) {
338 	case OVS_KEY_ATTR_PRIORITY:
339 		skb->priority = nla_get_u32(nested_attr);
340 		break;
341 
342 	case OVS_KEY_ATTR_ETHERNET:
343 		err = set_eth_addr(skb, nla_data(nested_attr));
344 		break;
345 
346 	case OVS_KEY_ATTR_IPV4:
347 		err = set_ipv4(skb, nla_data(nested_attr));
348 		break;
349 
350 	case OVS_KEY_ATTR_TCP:
351 		err = set_tcp(skb, nla_data(nested_attr));
352 		break;
353 
354 	case OVS_KEY_ATTR_UDP:
355 		err = set_udp(skb, nla_data(nested_attr));
356 		break;
357 	}
358 
359 	return err;
360 }
361 
362 /* Execute a list of actions against 'skb'. */
363 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
364 			const struct nlattr *attr, int len, bool keep_skb)
365 {
366 	/* Every output action needs a separate clone of 'skb', but the common
367 	 * case is just a single output action, so that doing a clone and
368 	 * then freeing the original skbuff is wasteful.  So the following code
369 	 * is slightly obscure just to avoid that. */
370 	int prev_port = -1;
371 	const struct nlattr *a;
372 	int rem;
373 
374 	for (a = attr, rem = len; rem > 0;
375 	     a = nla_next(a, &rem)) {
376 		int err = 0;
377 
378 		if (prev_port != -1) {
379 			do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
380 			prev_port = -1;
381 		}
382 
383 		switch (nla_type(a)) {
384 		case OVS_ACTION_ATTR_OUTPUT:
385 			prev_port = nla_get_u32(a);
386 			break;
387 
388 		case OVS_ACTION_ATTR_USERSPACE:
389 			output_userspace(dp, skb, a);
390 			break;
391 
392 		case OVS_ACTION_ATTR_PUSH_VLAN:
393 			err = push_vlan(skb, nla_data(a));
394 			if (unlikely(err)) /* skb already freed. */
395 				return err;
396 			break;
397 
398 		case OVS_ACTION_ATTR_POP_VLAN:
399 			err = pop_vlan(skb);
400 			break;
401 
402 		case OVS_ACTION_ATTR_SET:
403 			err = execute_set_action(skb, nla_data(a));
404 			break;
405 
406 		case OVS_ACTION_ATTR_SAMPLE:
407 			err = sample(dp, skb, a);
408 			break;
409 		}
410 
411 		if (unlikely(err)) {
412 			kfree_skb(skb);
413 			return err;
414 		}
415 	}
416 
417 	if (prev_port != -1) {
418 		if (keep_skb)
419 			skb = skb_clone(skb, GFP_ATOMIC);
420 
421 		do_output(dp, skb, prev_port);
422 	} else if (!keep_skb)
423 		consume_skb(skb);
424 
425 	return 0;
426 }
427 
428 /* Execute a list of actions against 'skb'. */
429 int ovs_execute_actions(struct datapath *dp, struct sk_buff *skb)
430 {
431 	struct sw_flow_actions *acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
432 
433 	return do_execute_actions(dp, skb, acts->actions,
434 					 acts->actions_len, false);
435 }
436